In [1]:
import numpy as np
import pandas as pd

import os
In [ ]:
# !pip install twython
In [2]:
import nltk
# nltk.download("vader_lexicon")
from nltk.sentiment.vader import SentimentIntensityAnalyzer as SIA

Data Preparation¶

In [3]:
vax_df = pd.read_csv("Datasets/vaccination_all_tweets.csv")
vax_df.head()
Out[3]:
id user_name user_location user_description user_created user_followers user_friends user_favourites user_verified date text hashtags source retweets favorites is_retweet
0 1340539111971516416 Rachel Roh La Crescenta-Montrose, CA Aggregator of Asian American news; scanning di... 2009-04-08 17:52:46 405 1692 3247 False 2020-12-20 06:06:44 Same folks said daikon paste could treat a cyt... ['PfizerBioNTech'] Twitter for Android 0 0 False
1 1338158543359250433 Albert Fong San Francisco, CA Marketing dude, tech geek, heavy metal & '80s ... 2009-09-21 15:27:30 834 666 178 False 2020-12-13 16:27:13 While the world has been on the wrong side of ... NaN Twitter Web App 1 1 False
2 1337858199140118533 eli🇱🇹🇪🇺👌 Your Bed heil, hydra 🖐☺ 2020-06-25 23:30:28 10 88 155 False 2020-12-12 20:33:45 #coronavirus #SputnikV #AstraZeneca #PfizerBio... ['coronavirus', 'SputnikV', 'AstraZeneca', 'Pf... Twitter for Android 0 0 False
3 1337855739918835717 Charles Adler Vancouver, BC - Canada Hosting "CharlesAdlerTonight" Global News Radi... 2008-09-10 11:28:53 49165 3933 21853 True 2020-12-12 20:23:59 Facts are immutable, Senator, even when you're... NaN Twitter Web App 446 2129 False
4 1337854064604966912 Citizen News Channel NaN Citizen News Channel bringing you an alternati... 2020-04-23 17:58:42 152 580 1473 False 2020-12-12 20:17:19 Explain to me again why we need a vaccine @Bor... ['whereareallthesickpeople', 'PfizerBioNTech'] Twitter for iPhone 0 0 False
In [4]:
import re
def clean(text):
    text = re.sub("https?://\S+|www\.\S+", "", text)
    text = re.sub(r"\s+", " ", text, flags=re.I)
    text = re.sub("\[.*?\]", "", text)
    text = re.sub("\n", "", text)
    text = re.sub("\w*\d\w*", "", text)
    text = re.sub("<.*?>+", "", text)
    return text
In [5]:
vax_df["text"] = vax_df["text"].apply(lambda x:clean(x))
In [6]:
sia = SIA()

scores = []
for i in range(len(vax_df["text"])):
    score = sia.polarity_scores(vax_df["text"][i])
    score = score["compound"]
    scores.append(score)
    
sentiment_score = []
for i in scores:
    if i >= 0.05:
        sentiment_score.append("Positive")
    elif i <= (-0.05):
        sentiment_score.append("Negative")
    else:
        sentiment_score.append("Neutral")

vax_df["sentiment"] = pd.Series(np.array(sentiment_score))
In [7]:
import string

def clean_text(text):
    text = str(text).lower()
    text = re.sub("[%s]" % re.escape(string.punctuation), "", text)
    
    return text

vax_df["text"] = vax_df["text"].apply(lambda x:clean_text(x))

vax_df["text"]
Out[7]:
0         same folks said daikon paste could treat a cyt...
1         while the world has been on the wrong side of ...
2         coronavirus sputnikv astrazeneca pfizerbiontec...
3         facts are immutable senator even when youre no...
4         explain to me again why we need a vaccine bori...
                                ...                        
228202     urban bengaluru covidvaccine availability for...
228203     bbmp bengaluru covidvaccine availability for ...
228204     urban bengaluru covidvaccine availability for...
228205    they promote their vaccines leaving out the st...
228206     urban bengaluru covidvaccine availability for...
Name: text, Length: 228207, dtype: object
In [8]:
vax_df = vax_df.dropna()

vax = vax_df.drop(["user_name","user_description","user_created","user_followers","user_friends","user_favourites","source","is_retweet"],axis=1)
In [9]:
vax.head()
Out[9]:
id user_location user_verified date text hashtags retweets favorites sentiment
0 1340539111971516416 La Crescenta-Montrose, CA False 2020-12-20 06:06:44 same folks said daikon paste could treat a cyt... ['PfizerBioNTech'] 0 0 Positive
2 1337858199140118533 Your Bed False 2020-12-12 20:33:45 coronavirus sputnikv astrazeneca pfizerbiontec... ['coronavirus', 'SputnikV', 'AstraZeneca', 'Pf... 0 0 Positive
6 1337851215875608579 Austria, Ukraine and Kosovo False 2020-12-12 20:06:00 it is a bit sad to claim the fame for success ... ['vaccination'] 0 4 Positive
9 1337842295857623042 Islamabad False 2020-12-12 19:30:33 covidvaccine states will start getting monday... ['CovidVaccine', 'COVID19Vaccine', 'US', 'paku... 0 0 Neutral
10 1337841934170255365 Turkey-Israel False 2020-12-12 19:29:07 while deaths are closing in on the mark milli... ['PfizerBioNTech', 'Vaccine'] 0 0 Neutral
In [10]:
totals = vax.groupby("sentiment").count()["text"].reset_index().sort_values(by="text", ascending=False)

totals.style.background_gradient(cmap="RdYlBu")
Out[10]:
  sentiment text
1 Neutral 53011
2 Positive 44993
0 Negative 18051

Training the model¶

In [11]:
from sklearn.feature_extraction.text import CountVectorizer
from nltk.tokenize import RegexpTokenizer

from nltk.tokenize import RegexpTokenizer
token = RegexpTokenizer(r"[a-zA-Z0-9]+")

cv = CountVectorizer(stop_words="english",ngram_range = (1,1),tokenizer = token.tokenize)
text_count = cv.fit_transform(vax["text"])

C:\Users\Tilda\anaconda3\Lib\site-packages\sklearn\feature_extraction\text.py:525: UserWarning: The parameter 'token_pattern' will not be used since 'tokenizer' is not None'
  warnings.warn(
In [12]:
print(text_count)

  (0, 19961)	1
  (0, 47425)	1
  (0, 13081)	1
  (0, 40532)	1
  (0, 56071)	1
  (0, 13018)	1
  (0, 52206)	1
  (0, 41230)	1
  (1, 41230)	1
  (1, 11625)	1
  (1, 51459)	1
  (1, 3724)	1
  (1, 34998)	1
  (1, 47143)	1
  (1, 58177)	1
  (1, 12483)	1
  (1, 61761)	1
  (2, 6198)	1
  (2, 47305)	1
  (2, 9944)	1
  (2, 18815)	1
  (2, 52532)	1
  (2, 58088)	1
  (2, 40631)	1
  (2, 10844)	1
  :	:
  (116051, 5058)	1
  (116052, 12246)	1
  (116052, 15610)	1
  (116052, 20376)	1
  (116052, 12329)	1
  (116052, 4119)	1
  (116052, 57644)	1
  (116052, 40039)	1
  (116052, 5470)	1
  (116052, 50404)	2
  (116053, 51459)	1
  (116053, 58177)	1
  (116053, 58425)	1
  (116053, 30769)	1
  (116053, 47129)	1
  (116053, 52352)	1
  (116053, 43347)	1
  (116054, 12246)	1
  (116054, 20376)	1
  (116054, 12329)	1
  (116054, 4119)	1
  (116054, 57644)	1
  (116054, 40039)	1
  (116054, 5470)	1
  (116054, 50404)	2
In [13]:
from sklearn.model_selection import train_test_split
X_train, X_test, Y_train, Y_test = train_test_split(text_count, vax["sentiment"], test_size=0.25, random_state=5)
In [14]:
print(X_train)

  (0, 47425)	1
  (0, 45053)	1
  (0, 50047)	1
  (0, 14277)	1
  (0, 27555)	1
  (0, 22295)	1
  (0, 30267)	1
  (0, 24470)	1
  (0, 43221)	1
  (0, 909)	1
  (0, 805)	1
  (0, 30599)	1
  (0, 62075)	1
  (0, 51658)	1
  (1, 34998)	1
  (1, 55643)	1
  (1, 50920)	1
  (1, 36075)	1
  (1, 6606)	1
  (1, 48537)	1
  (1, 56207)	1
  (1, 48453)	1
  (1, 44686)	1
  (1, 34389)	1
  (1, 33648)	1
  :	:
  (87038, 31712)	1
  (87038, 39088)	1
  (87038, 31710)	1
  (87039, 34998)	1
  (87039, 24663)	1
  (87039, 15610)	1
  (87039, 12891)	1
  (87039, 42395)	1
  (87039, 469)	1
  (87039, 35081)	1
  (87039, 6606)	1
  (87039, 19307)	1
  (87039, 9392)	1
  (87039, 47076)	1
  (87039, 30957)	1
  (87040, 41230)	1
  (87040, 58177)	1
  (87040, 57993)	1
  (87040, 12020)	1
  (87040, 55455)	1
  (87040, 31540)	1
  (87040, 8674)	1
  (87040, 60372)	1
  (87040, 22153)	1
  (87040, 20698)	1
In [15]:
from sklearn.naive_bayes import MultinomialNB

mnb = MultinomialNB()
mnb.fit(X_train, Y_train)
Out[15]:
MultinomialNB()
In a Jupyter environment, please rerun this cell to show the HTML representation or trust the notebook.
On GitHub, the HTML representation is unable to render, please try loading this page with nbviewer.org.
MultinomialNB()

Prediction¶

In [16]:
from sklearn import metrics
predictions = mnb.predict(X_test)
accuracy_score = metrics.accuracy_score(predictions, Y_test)

print(str("{:04.2f}".format(accuracy_score*100))+"%")

79.84%
In [17]:
from sklearn.naive_bayes import ComplementNB
from sklearn.naive_bayes import BernoulliNB

cnb = ComplementNB()
bnb = BernoulliNB()
In [18]:
from sklearn.feature_extraction.text import TfidfVectorizer
tfidf = TfidfVectorizer()
text_counts = tfidf.fit_transform(vax["text"])
In [19]:
x_train, x_test, y_train, y_test = train_test_split(text_counts, vax["sentiment"], test_size=0.25, random_state=5)

mnb.fit(x_train, y_train)
mnb_accuracy_score = metrics.accuracy_score(mnb.predict(x_test), y_test)
print("MNB accuracy score: "+str("{:04.2f}".format(mnb_accuracy_score*100))+"%")

cnb.fit(x_train, y_train)
cnb_accuracy_score = metrics.accuracy_score(cnb.predict(x_test), y_test)
print("CNB accuracy score: "+str("{:04.2f}".format(cnb_accuracy_score*100))+"%")

bnb.fit(x_train, y_train)
bnb_accuracy_score = metrics.accuracy_score(bnb.predict(x_test), y_test)
print("BNB accuracy score: "+str("{:04.2f}".format(bnb_accuracy_score*100))+"%")

MNB accuracy score: 77.35%
CNB accuracy score: 79.53%
BNB accuracy score: 80.46%

Data Analysis and Visualization¶

In [20]:
import matplotlib.pyplot as plt
import plotly.express as px
import plotly.graph_objects as go
import plotly.figure_factory as ff
from plotly.colors import n_colors
from plotly.subplots import make_subplots
In [63]:
#!pip install wordcloud
from wordcloud import WordCloud, STOPWORDS, ImageColorGenerator
In [22]:
df_pos = vax[vax["sentiment"] == "Positive"].reset_index()
df_neg = vax[vax["sentiment"] == "Negative"].reset_index()
df_neu = vax[vax["sentiment"] == "Neutral"].reset_index()

Distrubution of Sentiments Chart¶

In [23]:
percent_class=totals.text
labels= totals.sentiment

colors = ["#89c2d9","#52b788","#e5383b"]

my_pie,_,_ = plt.pie(percent_class,radius = 1.2,labels=labels,colors=colors,autopct="%.1f%%")

plt.setp(my_pie, width=0.7, edgecolor='white') 

plt.show()

Wordclouds¶

In [24]:
fig, (ax1, ax2, ax3) = plt.subplots(1, 3, figsize=[30, 15])

# Positive sentiment wordcloud
comment_words = ""
stopwords = set(STOPWORDS)

for val in df_pos.text:
    val = str(val)
    tokens = val.split()
    
    for i in range(len(tokens)):
        tokens[i] = tokens[i].lower()
        
    comment_words += " ".join(tokens)+" "
    
pos_wordcloud = WordCloud(width=900, height=900,
                          background_color="white",
                          colormap="Greens",
                          stopwords=stopwords,
                          min_font_size=10).generate(comment_words)
ax1.imshow(pos_wordcloud)
ax1.axis("off")
ax1.set_title("Positive Sentiment", fontsize=35)

# Negative sentiment wordcloud
comment_words = ""

for val in df_neg.text:
    val = str(val)
    tokens = val.split()
    
    for i in range(len(tokens)):
        tokens[i] = tokens[i].lower()
        
    comment_words += " ".join(tokens)+" "
    
neg_wordcloud = WordCloud(width=900, height=900,
                          background_color="white",
                          colormap="Reds",
                          stopwords=stopwords,
                          min_font_size=10).generate(comment_words)
ax2.imshow(neg_wordcloud)
ax2.axis("off")
ax2.set_title("Negative Sentiment", fontsize=35)

# Neutral sentiment wordcloud
comment_words = ""

for val in df_neu.text:
    val = str(val)
    tokens = val.split()
    
    for i in range(len(tokens)):
        tokens[i] = tokens[i].lower()
        
    comment_words += " ".join(tokens)+" "
    
neu_wordcloud = WordCloud(width=900, height=900,
                          background_color="white",
                          colormap="Blues",
                          stopwords=stopwords,
                          min_font_size=10).generate(comment_words)
ax3.imshow(neu_wordcloud)
ax3.axis("off")
ax3.set_title("Neutral Sentiment", fontsize=35)
Out[24]:
Text(0.5, 1.0, 'Neutral Sentiment')

N-grams¶

In [25]:
import unicodedata

def basic_clean(text):
    wnl = nltk.stem.WordNetLemmatizer()
    stopwords = nltk.corpus.stopwords.words("english") 
    text = (unicodedata.normalize("NFKD", text).encode("ascii", "ignore").decode("utf-8", "ignore").lower())
    words = re.sub(r'[^\w\s]', '', text).split()
    return [wnl.lemmatize(word) for word in words if word not in stopwords]
In [26]:
import nltk
# nltk.download("wordnet")
In [27]:
pos_tweets = basic_clean("".join(str(df_pos["text"].tolist())))

Positive Sentiment N-grams¶

In [28]:
positive_unigram = (pd.Series(nltk.ngrams(pos_tweets, 1)).value_counts())[:30]
positive_unigram.sort_values().plot.barh(color="#52b788", width=.9, figsize=(12, 8))

plt.title("Positive sentiment unigram")
Out[28]:
Text(0.5, 1.0, 'Positive sentiment unigram')
In [29]:
positive_bigram = (pd.Series(nltk.ngrams(pos_tweets, 2)).value_counts())[:30]
positive_bigram.sort_values().plot.barh(color="#399e6f", width=.9, figsize=(12, 8))

plt.title("Positive sentiment bigram")
Out[29]:
Text(0.5, 1.0, 'Positive sentiment bigram')
In [30]:
positive_trigram = (pd.Series(nltk.ngrams(pos_tweets, 3)).value_counts())[:30]
positive_trigram.sort_values().plot.barh(color="#2d8a5f", width=.9, figsize=(12, 8))

plt.title("Positive sentiment trigram")
Out[30]:
Text(0.5, 1.0, 'Positive sentiment trigram')
In [31]:
neg_tweets = basic_clean("".join(str(df_neg["text"].tolist())))

Negative Sentiment N-grams¶

In [32]:
negative_unigram = (pd.Series(nltk.ngrams(neg_tweets, 1)).value_counts())[:30]
negative_unigram.sort_values().plot.barh(color="#e5383b", width=.9, figsize=(12, 8))

plt.title("Negative sentiment unigram")
Out[32]:
Text(0.5, 1.0, 'Negative sentiment unigram')
In [33]:
negative_bigram = (pd.Series(nltk.ngrams(neg_tweets, 2)).value_counts())[:30]
negative_bigram.sort_values().plot.barh(color="#ba181b", width=.9, figsize=(12, 8))

plt.title("Negative sentiment bigram")
Out[33]:
Text(0.5, 1.0, 'Negative sentiment bigram')
In [34]:
negative_trigram = (pd.Series(nltk.ngrams(neg_tweets, 3)).value_counts())[:30]
negative_trigram.sort_values().plot.barh(color="#a4161a", width=.9, figsize=(12, 8))

plt.title("Negative sentiment trigram")
Out[34]:
Text(0.5, 1.0, 'Negative sentiment trigram')
In [35]:
neu_tweets = basic_clean("".join(str(df_neu["text"].tolist())))

Neutral Sentiment N-grams¶

In [36]:
neutral_unigram = (pd.Series(nltk.ngrams(neu_tweets, 1)).value_counts())[:30]
neutral_unigram.sort_values().plot.barh(color="#89c2d9", width=.9, figsize=(12, 8))

plt.title("Neutral sentiment unigram")
Out[36]:
Text(0.5, 1.0, 'Neutral sentiment unigram')
In [37]:
neutral_bigram = (pd.Series(nltk.ngrams(neu_tweets, 2)).value_counts())[:30]
neutral_bigram.sort_values().plot.barh(color="#61a5c2", width=.9, figsize=(12, 8))

plt.title("Neutral sentiment bigram")
Out[37]:
Text(0.5, 1.0, 'Neutral sentiment bigram')
In [38]:
neutral_trigram = (pd.Series(nltk.ngrams(neu_tweets, 3)).value_counts())[:30]
neutral_trigram.sort_values().plot.barh(color="#468faf", width=.9, figsize=(12, 8))

plt.title("Neutral sentiment trigram")
Out[38]:
Text(0.5, 1.0, 'Neutral sentiment trigram')

Variance with Time for Vaccines¶

In [39]:
import time
import datetime

vax["date"] = pd.to_datetime(vax["date"], errors="coerce").dt.date

vax.head()
Out[39]:
id user_location user_verified date text hashtags retweets favorites sentiment
0 1340539111971516416 La Crescenta-Montrose, CA False 2020-12-20 same folks said daikon paste could treat a cyt... ['PfizerBioNTech'] 0 0 Positive
2 1337858199140118533 Your Bed False 2020-12-12 coronavirus sputnikv astrazeneca pfizerbiontec... ['coronavirus', 'SputnikV', 'AstraZeneca', 'Pf... 0 0 Positive
6 1337851215875608579 Austria, Ukraine and Kosovo False 2020-12-12 it is a bit sad to claim the fame for success ... ['vaccination'] 0 4 Positive
9 1337842295857623042 Islamabad False 2020-12-12 covidvaccine states will start getting monday... ['CovidVaccine', 'COVID19Vaccine', 'US', 'paku... 0 0 Neutral
10 1337841934170255365 Turkey-Israel False 2020-12-12 while deaths are closing in on the mark milli... ['PfizerBioNTech', 'Vaccine'] 0 0 Neutral
In [40]:
vax["sentiment"] = vax["sentiment"].map( {'Positive':1 , 'Neutral':0, 'Negative':-1} )
In [41]:
all_vax = ["covaxin", "sinopharm", "sinovac", "moderna", "pfizer", "biontech", "oxford", "astrazeneca", "sputnik"]

vax_sentiment = pd.DataFrame()
vax_sentiment["Vaccine"]=all_vax
sentiments=list()
In [42]:
import seaborn as sns

def vax_time_variance(vacc) :
    if vacc=="all":
        vacc=vax
    else:   
        vacc=vax[vax["text"].str.lower().str.contains(vacc)]
        sentiments.append(vacc.sentiment.mean())

    temp=pd.DataFrame()
    temp["date"] = sorted(vacc["date"].unique())
    senti=list()

    for date in temp["date"]:
        senti.append(vacc[vacc["date"]==date].sentiment.mean())

    temp["sentiment"]=senti
    
    fg,axs=plt.subplots(figsize=(15,7))
    sns.lineplot(ax=axs,x="date",y="sentiment", data= temp)
    axs.set_xlabel("Time",size=16)
    axs.set_ylabel("sentiment",size=16)
    axs.set_title("Variance of Sentiment with Time",size=24)
    axs.grid()

All Vaccines¶

In [43]:
vax_time_variance("all")

Covaxin¶

In [44]:
vax_time_variance("covaxin")

Sinopharm¶

In [45]:
vax_time_variance("sinopharm")

Sinovac¶

In [46]:
vax_time_variance("sinovac")

Moderna¶

In [47]:
vax_time_variance("moderna")

Pfizer¶

In [48]:
vax_time_variance("pfizer")

Biontech¶

In [49]:
vax_time_variance("biontech")

Oxford¶

In [50]:
vax_time_variance("oxford")

Astrazeneca¶

In [51]:
vax_time_variance("astrazeneca")

Sputnik¶

In [52]:
vax_time_variance("sputnik")

Variance with Time in Different Countries¶

In [53]:
def country_time_variance(country) :
    count=vax[vax["user_location"].str.lower().str.contains(country)]

    temp=pd.DataFrame()
    temp["date"] = sorted(count["date"].unique())
    senti=list()

    for date in temp["date"]:
        senti.append(count[count["date"]==date].sentiment.mean())

    temp["sentiment"]=senti
    
    fg,axs=plt.subplots(figsize=(15,7))
    sns.lineplot(ax=axs,x='date',y="sentiment", data= temp)
    axs.set_xlabel("Time",size=16)
    axs.set_ylabel("sentiment",size=16)
    axs.set_title("Variance of Sentiment with Time",size=24)
    axs.grid()

Sweden¶

In [54]:
country_time_variance("sweden")

USA¶

In [55]:
country_time_variance("usa")

UK¶

In [56]:
country_time_variance("uk")

Norway¶

In [57]:
country_time_variance("norway")

Finland¶

In [58]:
country_time_variance("finland")

Denmark¶

In [59]:
country_time_variance("denmark")

Mean Sentiments¶

Vaccines¶

In [60]:
vax_sentiment["sentiment"]=sentiments

fg,axs=plt.subplots(figsize=(15,7))
sns.barplot(ax=axs,x="Vaccine",y="sentiment",data=vax_sentiment)
axs.set_xlabel("Vaccines",size=16)
axs.set_ylabel("Mean Sentiment",size=16)
axs.set_title("Mean Sentiment of Vaccines",size=24)
Out[60]:
Text(0.5, 1.0, 'Mean Sentiment of Vaccines')
In [61]:
country_sentiment=pd.DataFrame()
countries=["sweden","usa","uk","norway","finland","denmark","germany","canada","russia","italy","australia","france","argentina","uae","israel","mexico","japan", "india", "china"]
country_sentiment['countries']=countries
senti=list()

for country in countries :
    senti.append(vax[vax["user_location"].str.lower().str.contains(country)].sentiment.mean())
    
country_sentiment["sentiment"]=senti

Countries¶

In [62]:
fg,axs=plt.subplots(figsize=(15,10))
sns.barplot(ax=axs,x="countries",y="sentiment",data=country_sentiment)
axs.set_xlabel("Countries",size=16)
axs.set_ylabel("Mean Sentiment",size=16)
axs.set_title("Mean Sentiment of Countries",size=24)
axs.grid()

CONCLUSION¶

Data used for this notebook

This analysis only shows sentiments on Twitter during a limited time period (2020-12-12 - 2021-11-23). This notebook only represents sentiments on Twitter during this time, and does not reflect all social media platforms nor possible change in sentiments during other time periods.

As shown in the Distrubution of Sentiments Chart, there is an uneven distribution of sentiments, with neutral taking the highest portion of the total percentage. Positive sentiments follow closely after, while negative sentiments hold a significantly lower percentage.

The Positive Sentiment N grams seem to mainly contain Tweets about the availability and approval of Covid vaccines, possibly particularly in rural areas since mentions of rural and urban is present in all positive N-grams. India is also frequently mentioned. Another common theme in the positive sentiment tweets is people sharing that they got vaccinated.

Negative Sentiment N grams show a big number of Tweets about Canada and India. These N-grams seem to mainly contain concerns about health, death and side effects of the vaccines. They also seem to contain reports of death cases. Mentions of lockdowns and staying at home is unique to the trigram in this category.

Similar to previous N-grams, Neutral Sentiment N grams contain many Tweets about India and Canada, but also Russia and China. The Tweets in this category seem to mainly report the spread of the virus. Like the positive sentiment N-grams, there are many who post about getting vaccinated.

As seen in both the graphs for variance with time for vaccines and variance with time for countries, the variance in sentiments are quite erratic, some more so than others. Most of these graphs show big spikes between positive and negative sentiments, and rarely any stable values. It is worth pointing out that all of these graphs only reflect early 2021 to Nov 2021.

Lastly, the Mean Sentiments of Countries show a great variance between sentiments of different countries. One that stands out in particular is Canada, which scores significantly lower than other countries in the chart. This may correlate with the frequent mentions of Canada in the negative sentiment N-grams.